Introduction to Dynamic Workflows

Dynamic workflows are now officially available in Copilot CLI, the GitHub Copilot app, and the GitHub Copilot SDK, bringing a new level of reliability and observability to complex, multi-agent tasks as announced via the GitHub Changelog.

Unlike standard prompt interactions, a dynamic workflow functions as a programmed routine that dictates how a task is carried out. It merges automated procedural steps with the analytical capabilities of one or more agents. These steps can execute consecutively, concurrently, or through a combination of both execution methods.

How Code-Defined Orchestration Works

The operational steps, the precise moments to engage AI agents, and the methods for utilizing their outputs are all defined directly within code. Meanwhile, agents manage the specific tasks that require deep analysis or judgment.

Because the program resides inside a GitHub Copilot extension, it gains direct access to Copilot’s powerful extensibility APIs. This architecture allows developers to run commands, utilize various tools, call external services, partition goals into parallel tasks, and pass structured findings from one stage to the next.

Additionally, workflows can pause at designated checkpoints to await user input or review before resuming, offering fine-grained control over complex processes. Detailed information on how these orchestrations function can be found in our docs about dynamic workflows.

Use Cases and Practical Applications

Developers can design dynamic workflows to handle repetitive or multi-stage processes that require clear checks and structural limits. Common scenarios include investigating service incidents by collecting telemetry logs, assigning independent agents to analyze separate systems, and merging structured findings into a root-cause report.

Other practical applications include running release checks that pause for human review upon failure, reviewing multiple changed files in a pull request simultaneously, and scanning large codebases across several directories to spot missing tests or deprecated API usages.

Authoring and Getting Started

Users have the option to author dynamic workflows manually or have Copilot write them automatically. Copilot includes built-in authoring guidance to help developers understand the underlying mechanics or generate complete workflows from scratch.

For step-by-step instructions on setting up your first routine, you can consult the official guide on Creating a dynamic workflow.

Availability and Access Requirements

Dynamic workflows are accessible across all GitHub Copilot plans. In the GitHub Copilot desktop app, the feature is always available without requiring any initial setup.

For users operating within the Copilot CLI, experimental features must be explicitly enabled by launching the CLI with the --experimental command-line flag or by executing the /experimental on command during an active interactive session. Developers can also update their CLI environment by typing /update.

Because the feature is currently in public preview, it remains subject to change as the team gathers user feedback through the Copilot CLI and the GitHub Community.

Sources

• GitHub ChangelogDynamic workflows in Copilot CLI and the Copilot app